An Analysis of the Characteristics and Detection Techniques for

Credit Card Fraudulent Transactions

Haojun Shi

Lee Shau Kee School of Business and Administration, Hong Kong Metropolitan University, Hong Kong, China

Keywords: Credit Card Fraud, Machine Learning, Deep Learning, Transaction Security, Data Imbalance.

Abstract: With the rapid growth of the modern economy, credit cards have become more and more essential for daily

transactions. At the same time, the rise in fraudulent activities poses significant financial risks, making it of

critical importance for cardholders and merchants to implement effective detection measures. This paper

reviews various machine learning and deep learning techniques aimed at accurately identifying fraudulent

credit card transactions. It covers key aspects such as technology analysis, detailed feature extraction,

algorithm selection, and the overall effectiveness of these methods. The discussion includes the strengths and

weaknesses of different algorithms, particularly in handling imbalanced data and complex fraud patterns.

Ultimately, this research offers insights for institutions seeking to enhance security, mitigate financial risks,

and advance detection technology with innovative solutions. In the future, the research should focus on

improving scalability and real-time detection capabilities to better address evolving fraud strategies,

ultimately contributing to more secure financial ecosystems.

1 INTRODUCTION

As a non-cash credit tool provided by financial

institutions, a credit card plays an indispensable role

in the modern economic life. It allows cardholders to

make purchases without having to pay cash

immediately and to complete repayments within

subsequent billing cycles, promoting convenience

and flexibility. Unlike debit cards, which are charged

directly from the user's account, credit card

transactions are limited by a preset credit limit,

determined by the card issuer based on an assessment

of the cardholder's credit status (Bertaut & Haliassos,

2006).

Credit card transactions are a complex and

sophisticated ecosystem typically involving five core

entities: cardholders, merchants, card issuers, credit

card organizations, and acquiring agencies. The card

issuer is responsible for issuing credit cards to

cardholders and providing related services. The

merchant processes the payment made by the

customer using the credit card through the acquiring

agency. Credit card organizations act as clearing

centers for transactions, ensuring accurate fund

transfers between parties. After the cardholder

https://orcid.org/0009-0003-2765-4696

completes the consumption, the merchant submits the

transaction details to the bank for approval, and once

the approval is passed, the bank will pay the merchant

on behalf of the cardholder while reducing the credit

limit of the cardholder.

Given the central role of credit card transactions

in the modern economy, transaction detection is

particularly important (Chu et al., 2023). Effective

transaction detection mechanisms detect and prevent

fraud promptly, protecting the property of

cardholders, merchants, and financial institutions.

Through advanced technologies such as big data

analysis and artificial intelligence, transaction

patterns can be deeply mined and abnormal

transaction characteristics can be identified to

effectively prevent risks such as credit card theft and

fake transactions (Varmedja et al., 2019). In addition,

transaction detection helps improve the quality of

customer service and provides cardholders with a

more convenient and secure payment experience by

optimizing the transaction process and improving

processing efficiency.

This paper starts with the characteristic analysis

of credit card fraudulent transactions and discusses

the detection methods based on machine learning and

532

Shi, H.

An Analysis of the Characteristics and Detection Techniques for Credit Card Fraudulent Transactions.

DOI: 10.5220/0013270000004568

In Proceedings of the 1st International Conference on E-commerce and Artiﬁcial Intelligence (ECAI 2024), pages 532-535

ISBN: 978-989-758-726-9

deep learning in detail. Through a systematic review

of the application of these technologies in credit card

fraud detection, the purpose is to reveal their

effectiveness and existing technical limitations and

provide references for future research directions. It is

hoped that this analysis will serve as a useful resource

for researchers and practitioners, encouraging further

exploration and refinement of fraud detection

methods in future studies.

2 FRAUD CHARACTERISTICS

The development of credit card transactions began in

the mid-20th century. In 1950, Diners Club issued the

first modern credit card for restaurant and travel

purchases. In 1958, American Express and banks

issued credit cards, which gradually became popular

in the retail industry. In the 1960s, advances in

computer processing technology led to the

widespread use of credit cards. Subsequently, the

development of magnetic strips, chip technology, and

online payments made credit card transactions more

secure and convenient.

Fraudulent credit card transactions often exhibit

multiple significant characteristics, including but not

limited to abnormal transaction patterns, abnormal

transaction time, abnormal transaction location,

abnormal transaction type, and abnormal login

information. The abnormal trading pattern is

embodied in two aspects. The first is the unusually

high amount of spending, which far exceeds the

cardholder's previous transaction records and is

highly unusual. This is followed by frequent small

test transactions, which may be used as a prelude to

test the validity of the card and then gradually

increase the transaction amount for fraudulent

purposes. A transaction time anomaly refers to the

period in which a transaction occurs that does not

match the cardholder's regular activity pattern and

can be replaced by a discrete period (year, month,

day, minute, and second) with a more easily analyzed

period label (such as morning, noon, or evening)

using machine learning technology. If the transaction

occurs during a very long period, such as late at night

or during holidays, and frequent trading activities

occur in a short period, such as a large number of

transactions in a few minutes or hours, it may be a

fraudulent transaction (Lu & Wung, 2015). The

transaction location anomaly is reflected in the fact

that the cardholder of the transaction goes to a place

that is infrequent or far away from his residence,

especially the transaction that spans multiple

countries or cities in a short period, which is contrary

to normal consumption habits.

Abnormal transaction type refers to the

transaction behavior that is inconsistent with the

cardholder's daily consumption habits, such as

sudden large purchases of virtual products, high-risk

products, luxury goods, or repeated purchases of the

same goods in a short period, which are not in line

with the normal consumption logic. Finally, abnormal

to log in with the wrong password, using false

personal information to make transactions, and

initiating transactions from unknown or high-risk

new devices and IP addresses. These behavior

patterns are highly abnormal, and there is a high

probability that they are fraudulent transactions.

3 APPLICATION OF MACHINE

LEARNING

Through the use of data and algorithms, machine

learning is a branch of computer science that enables

computers to perform better and learn continuously

from their data in order to enhance data processing

accuracy. The creation and study of statistical

algorithms that can learn from data and generalize to

invisible data to carry out tasks without explicit

instructions is known as machine learning, a branch

of artificial intelligence research. Chen et al. (2018)

used the machine learning technique of integrated

trees and neural networks to synthesize the literature

on credit card fraud and demonstrate the validity of

the machine learning model for credit card fraud.

Support vector machines (SVM), K-nearest neighbor

(KNN), and artificial neural networks (ANNs) were

the methods Asha and KR (2021) employed to detect

fraud; the artificial neural network's accuracy rate was

an astounding 100%. It is evident that machine

learning techniques are quite beneficial when it

comes to credit card fraud detection.

Because there are significantly fewer fraudulent

transactions than there are legitimate transactions, the

distribution of data in this standard binary

classification problem of credit card transactions is

imbalanced, making it more challenging for machine

learning to identify outliers. The machine learning

model's accuracy won't be impacted even if it

misinterprets fraudulent transactions as legitimate

ones. Therefore, how to ensure the stability of data

and construct a stable model is the direction of

research.

An Analysis of the Characteristics and Detection Techniques for Credit Card Fraudulent Transactions

533

In a recent study, the training set was balanced and

the ratio of positive to negative samples was adjusted

to 1:1 using six oversampling techniques, including

Synthetic Minority Over-sampling Technique

(SMOTE) (Huang, 2023). Awoyemi et al. (2017)

used downsampling and oversampling methods and

found that K-proximity performed better than naive

Bayes and logistic regression algorithms on

unbalanced data problems. Duman (2013) conducted

experiments with different proportions of data to

downsample transaction data. Yang (2021) used a

generative adversarial network (GAN) to alleviate the

imbalance of data and combined GAN and

convolutional neural network (CNN) algorithms to

demonstrate the superiority of model checking. Liao

(2022) used a random forest (RF) classifier to build a

detection model with low sensitivity to unbalanced

data sets. These techniques can significantly enhance

the capacity to identify credit card transactions.

4 APPLICATIONS OF DEEP

LEARNING

Deep learning is a branch of machine learning that

processes complex data by simulating artificial neural

networks. This process involves layers of sampling,

layers of representation, and learning algorithms for

data representation. Compared with other machine

learning algorithms, it is more inclined to the goal of

artificial intelligence, and common models include

CNN and recurrent neural networks (RNNs). Inspired

by real neurons, Warren McCulloch and Walter Pitts

created the first artificial neurons in the 1940s and

1960s. Frank Rosenblatt created Perceptron, an early

neural network model that could classify binary data.

It lays a foundation for the future fraud detection class

binary classification problem. In the 1970s-1990s,

Ronald J. Williams popularized backpropagation, a

method for training multi-layer neural networks that

made it possible to train more complex networks;

however, neural networks declined due to

computational limitations and the rise of SVMs and

other algorithms. New algorithms and improvements

in processing power have resulted from this. As a

result, deep learning was used in many domains in the

2010s, and methods like CNN, RNN, and GAN

became indispensable instruments.

According to the research, the combination of

RFE and oversampling, such as SMOTE, is better

than the traditional method in dealing with the high

imbalance of data, while the traditional method often

treats the transaction as an isolated event, to lack of

mining the interactive information, which largely

satisfies the imbalance of credit card fraud data (Liao

， 2022). In the aspect of model introduction, a deep

belief network (DBN) model is used, whose function

is to extract and classify fraudulent transaction

features. The model is trained using a variety of

classifiers, such as gradient lift trees (GBDT),

extreme gradient boosting (XGBoost), and mesh

search methods, to optimize the model parameters.

Huang (2023) may be able to help us further

decompress the concept of common P-values by

controlling marginal P-values and calibrating P-

values to increase discrimination since marginal P-

values have the advantage of error discovery rate

control. It is concluded that the Adaboost model is the

best; light, GBDT, and XGBoost are second. Xue

(2023) believes that compared with the traditional

model, the stochastic integrated deep learning model

has more advantages in data processing and

recognition ability, especially in identifying rare

fraudulent transaction samples, and it is not suitable

for big data samples such as credit card detection. She

proposed the innovative and improved Bagging and

Borderline-SMOTE algorithms. Can improve the

accuracy of the model. Deep learning techniques like

ANN, CNN, Gated Recurrent Units (GRU), and Long

Short-Term Memory (LSTM) each have their own

advantages in terms of algorithms. LSTM is

frequently utilized for anomaly detection in the

dynamic changes of complicated networks. Wang

(2019) proposed an anomaly detection method for a

dynamic graph model based on LSTM to effectively

characterize the changes in dynamic graph structure.

It is precisely because LSTM is capable of processing

time series data to capture the timing of transactions

in credit card transaction detection. By combining

LSTM with the feature fusion method, transaction

information of neighborhood nodes can be obtained

through random walks and dynamic features can be

added to the LSTM model to better capture the

interrelationship between transactions and potential

abnormal transaction patterns. GRU is similar to

LSTM, but its structure is simpler and the operation

speed is faster. The researchs can use GRU to model

and learn the historical transactions of users to find

abnormal transactions that deviate from the normal

pattern. In addition, CNN is a neural network model

that can extract spatial correlations of data. CNN is

less effective in the field of credit card detection than

LSTM and GRU because it is better suited for

processing static image data and has comparatively

poor time dependence and sequence history

processing capabilities. ANN's results are also

relatively simple, and it is suitable for dealing with

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

534

large-scale nonlinear problems. However, like CNN,

Ann's processing of time dependence and sequence

features is unstable, and it is easily affected by

gradient disappearance or explosion, so its effect is

far less than CNN's.

5 LIMITATION AND PROSPECT

Credit card fraud detection has become significantly

more effective due to the rapid advancements in

machine learning. There are still a lot of restrictions,

though. The percentage of fraudulent transactions

generally makes up very little of all transactions. A

significant disparity between positive and negative

samples might impact the model's training effect,

leading to the majority of learning models exhibiting

a tendency to forecast typical transactions. Secondly,

fraud changes too quickly and is too diversified.

Fraudsters constantly update their fraud methods,

which increases the difficulty of detection.

Traditional rule-based methods make it difficult to

deal with new fraud behaviors, and the rules need to

be updated and adjusted constantly. To increase the

precision and resilience of the detection model, future

research should incorporate multi-source data, which

can include information from several sources like

social media, bank transaction records, and

geographic location data. Algorithms that are

adaptive to detect fresh fraud patterns and improve

the ability to recognize new fraud behaviors can also

be developed. Additionally, real-time detection

efficiency can be increased by optimizing the

decision-making process and detection method using

reinforcement learning.

6 CONCLUSIONS

Finding and stopping fraudulent credit card

transactions is crucial to lowering financial risks. This

study examines machine learning and deep learning-

based detection techniques and evaluates the

capabilities and drawbacks of several algorithms for

handling fraudulent transactions. Although these

techniques have significantly increased the

effectiveness of detection, there is still much work to

be done to address the issues of data imbalance and

fraud diversity. In order to improve the model's

capacity to identify novel fraud patterns, integrate

data from multiple sources, and improve the system's

real-time performance, future research should

concentrate on these areas. This will enable financial

institutions to implement a more dependable risk

prevention strategy. Generally speaking, credit card

fraud cannot be prevented, and the transaction

detection model must be updated on a regular basis.

In order to identify irregularities and stop credit card

fraud, cardholders must cultivate a sense of security

and routinely organize the transaction flow.

REFERENCES

Asha, R. B., KR, S. K., 2021. Credit card fraud detection

using artificial neural network. Global Transitions

Proceedings, 2(1), 35-41.

Awoyemi, J. O., Adetunmbi, A. O., Oluwadare, S. A., 2017.

Credit card fraud detection using machine learning

techniques: A comparative analysis. In 2017

international conference on computing networking and

informatics (ICCNI) (pp. 1-9). IEEE.

Bertaut, C. C., Haliassos, M. (2006). Credit cards: facts and

theories.

Chu, Y. B., Lim, Z. M., Keane, B., Kong, P. H., Elkilany,

A. R., Abusetta, O. H., 2023. Credit Card Fraud

Detection on Original European Credit Card Holder

Dataset Using Ensemble Machine Learning Technique.

Journal of Cyber Security, 5, 33-46.

Duman, E., Buyukkaya, A., Elikucuk, I., 2013. A novel and

successful credit card fraud detection system

implemented in a turkish bank. In 2013 IEEE 13th

International Conference on Data Mining Workshops

(pp. 162-171). IEEE.

Huang, A., 2023. Research on credit card fraud detection

based on machine learning and conformal p-values

(Master's thesis, Huazhong University of Science and

Technology).

Lu, H. P., Wung, Y. S., 2021. Applying transaction cost

theory and push-pull-mooring model to investigate

mobile payment switching behaviors with well-

established traditional financial infrastructure. Journal

of theoretical and applied electronic commerce

research, 16(2), 1-21.

Liao, Q., 2022. Research on anomaly detection in Bitcoin

transactions (Master's thesis, People's Public Security

University of China).

Varmedja, D., Karanovic, M., Sladojevic, S., Arsenovic,

M., Anderla, A., 2019. Credit card fraud detection-

machine learning methods. In 2019 18th International

Symposium INFOTEH-JAHORINA (INFOTEH) (pp.

1-5). IEEE.

Wang, K., Chen, D., 2019. Research on anomaly detection

algorithm for dynamic graph model based on LSTM.

Computer Engineering and Applications, (05), 76-82.

Xue, X., 2023. Design of random ensemble learning

algorithm and credit card fraud detection (Master's

thesis, Yunnan University of Finance and Economics).

Yang, H., 2021. Research on credit card detection and

recognition based on GAN-CNN imbalanced

classification algorithm (Master's thesis, Shantou

University).

An Analysis of the Characteristics and Detection Techniques for Credit Card Fraudulent Transactions

535